Improved interfacial li-ion transport in composite polymer electrolytes via surface modification of LLZO

Materials

PEO (M_v~600 kDa), LiTFSI (99.95% metals basis), acetonitrile (ACN, 99.8%, anhydrous), APTES (98%), GPS (98%), toluene (99.8%, anhydrous), lithium nitrate (LiNO₃), lanthanum(III) nitrate hydrate [La(NO₃)₃•xH₂O], zirconium(IV) oxynitrate hydrate [ZrO(NO₃)₂•xH₂O], aluminum nitrate nonahydrate [Al(NO₃)₃•9H₂O], and dimethylformamide (DMF) were purchased from Sigma-Aldrich. Methanol (MeOH, HPLC grade), acetic acid (AcOH, glacial, Certified ACS), and polyvinylpyrrolidone (PVP, average MW = 1.3 MDa) were purchased from Fisher Scientific. Millipore water was used as a pure water source. Al-doped LLZO NPs (nominal composition of Li_6.25Al_0.25La₃Zr₂O₁₂) with a size of D₅₀ = 500 nm were from Ampcera. Al-doped LLZO NFs with the same composition and 200-300 nm diameter were prepared via electrospinning and annealing according to our previous work^[10]. The chemical composition of LLZO NPs and LLZO NFs was characterized by inductively coupled plasma optical emission spectroscopy (ICP-OES, Agilent 5110) [Supplementary Table 1]. Ta-doped LLZO sintered pellets prepared via solid-state reaction^[21] were generously provided by Dr. Chih-Long Tsai. Ionic conductivity of Ta-doped LLZO pellets was 5.87 10^-4 S cm^-1 and electronic conductivity is 5.11 × 10^-10 S cm^-1 at 25 °C, with the Li-ion transfer number 0.9976. N₂ absorption Brunauer-Emmett-Teller (BET) and Krypton BET measurements did not give any result due to the very low surface area of the pellets, even though we used many pellets to fill up the experimental glass container. For the relative density, by Archimedes’ method, we measured theoretical density 5.35 g cm^-3, relative density 91.6% ± 0.4%, open porosity 0.2% ± 0.15% and close porosity 8.2% ± 0.16%.

Lithium foil electrodes [Li, 250 µm thickness, Micro Technology Innovation Korea (MTI Korea)] were polished with a steel brush and lint-free cloth to remove surface contaminants. Au@Cu electrodes were prepared by first cleaning 1.0 cm² copper foil (Cu, 6 µm thick, MSE Supplies) discs with dilute HCl and acetone, followed by sputtering 20 nm of gold at a rate of 1.0 Å/s (DST1-170 sputter coater, Element Pi). Isotopically enriched ⁶Li metal (95 atom %, Sigma-Aldrich) was cut and pressed between steel plates before polishing, followed by typical cell assembly with electrode areas of 0.7-1.0 cm².

Composite electrolytes

CPEs with 0, 10 and 50 wt % LLZO NPs were prepared by mixing PEO, LiTFSI, and LLZO NPs in acetonitrile. Briefly, LiTFSI and LLZO were first mixed by ultrasonication for 30 min, and then PEO was added into the solution followed by 30 min mixing with a dual-asymmetric centrifuge speed mixer (FlackTek SpeedMixer®). The PEO concentration was kept at 6 wt % (EO:Li ratio = 15:1), and the LLZO mass ratio was respected to that of PEO. After preparation, the viscous slurry was coated onto a polytetrafluoroethylene (PTFE) film (McMaster-Carr) via slot-die coating (FOM vectorSC, FOM Technology) and dried in ambient to yield homogeneous membranes of 50-60 µm thickness with ±2 µm variation in thickness across the membrane. Good ductility and relatively uniform fiber distribution in the polymer matrix were observed for these membranes [Supplementary Figure 1]. The authors also found that thinner membranes with a desired thickness of less than 20 µm and uniform fiber distribution can be obtained by optimizing slurry ink rheological properties and coating conditions such as slurry injection rate, coating gap, and web moving speed. The dry membranes were punched into 1.6 cm² area discs and into a glovebox for cell assembly. PEO-LiTFSI samples (0 wt % LLZO NP) were processed in an identical process but did not include LLZO nanomaterials.

For silane modification, LLZO was thermally treated at 620 °C for 30 min in argon to remove surface carbonates. Silane modification of LLZO was carried out by dispersing 1 g of LLZO NPs or 0.1 g of LLZO NFs in 80 mL toluene in a round-bottom flask inside the glovebox. After adding 3 ml of APTES or GPS, the mixture was stirred and refluxed under nitrogen for 24 h. The functionalized particles were centrifuged and washed with ethanol several times before drying to remove excess silane reagent. For the preparation of Al₂O₃-coated LLZO NPs, atomic layer deposition (ALD) of Al₂O₃ on LLZO NPs was performed in accordance with our previous work^[10] for a Al₂O₃ thickness of ~10 nm. LLZO samples are noted as Neat LLZO, APTES@LLZO, GPS@LLZO, and Al₂O₃ for unmodified, APTES-functionalized, GPS-functionalized, and Al₂O₃-coated LLZO materials, respectively.

Electrochemical measurements

All electrochemical measurements were performed on a Biologic VSP300 potentiostat in a temperature-controlled chamber (ESPEC). Composite electrolytes were sandwiched between 0.9 cm² Li foil electrodes that were freshly polished. These were then assembled in 2032 coin cells with a 1.0 mm stainless steel spacer and wave spring; all cell parts were 304 stainless steel. Experiments were performed in triplicate with three cells of each composite electrolyte.

For trilayer electrolyte measurements, Ta-doped LLZO pellets were annealed in a muffle furnace inside an Ar-filled glovebox to remove surface contaminants^[22-24]. The samples were placed in a MgO crucible with a lid and heated at 250 °C then 500 °C, with 5 °C/min ramps and one hour dwells each step before cooling naturally overnight. For APTES@LLZO pellets, the annealed LLZO was soaked in a 3 wt % solution of APTES in ACN inside the glovebox for several hours to mitigate any air exposure from the procedure required for LLZO NP functionalization, then rinsed thoroughly with ACN and dried. While this method differs from LLZO NP functionalization, it is still suitable for functionalization of the flat pellet surface compared to the more rigorous, longer reflux in toluene needed for high surface area NPs. After surface preparation, PEO-LiTFSI was drop-cast from a dilute ACN solution onto one side of the pellet, dried at 60 °C under vacuum for 30 min, and the same was repeated on the other pellet side. The total electrolyte thickness was measured after each drying step to determine PEO-LiTFSI thickness. Coin cells with 20 nm Au@Cu electrodes were prepared as above with a 0.5 mm spacer.

Cells were ramped from 5 to 70 °C at 5° C increments every 30 minutes, then ramped down and allowed to equilibrate at 60 °C before subsequent measurements. Electrochemical impedance (EIS) measurements were performed with a 7 MHz-1 or 0.1 Hz frequency range, 0.0 V vs. open circuit DC bias, 10 mV amplitude AC sine wave, and 20 points per decade sampling. Chronoamperometry was performed at +50 mV bias for 1,000 s with 1 mV/1 s sampling intervals to capture accurate currents. Galvanostatic cycling was performed in 10 µA/cm², 10 min steps with 40 min total rest time and EIS measurements in between each step. Transference numbers were estimated as the Bruce-Vincent current fraction^[25]:

where i₀ and i_ss are the initial and steady-state currents from chronoamperometry, ΔV is the applied bias, and R_{Anode-Int,0/ss} is the medium frequency resistance from EIS measurements before and after chronoamperometry. The apparent Arrhenius activation energy was determined for narrow temperature ranges via:

where σ is the calculated conductivity from impedance measurements, E_a is the Arrhenius activation energy, R is the gas constant, and T is temperature.

Materials characterization

Fourier transform infrared spectroscopy (FT-IR) was performed using a Bruker Vertex 70 with an Attenuated Total Reflection (ATR) accessory. Raman microscopy was performed on a Renishaw inVia system with 633 nm laser at 5 mW and a 50x objective (NA = 0.5).

Solid-state NMR spectroscopy was performed on a Bruker Avance III wide-bore 400 MHz spectrometer under a field of 9.5 T. Samples were packed into a 1.9 mm ZrO₂ rotor with polyimide caps. All spectra were acquired while spinning at 20 kHz. ⁷Li 1D spectra were acquired using the Larmor frequency of 155.5 MHz and a 90^o pulse time of 0.9 µs with a corresponding power level of 90 W. Recycle delay times were optimized for each sample and varied between 20 - 40 s. ⁶Li 1D spectra were acquired using the Larmor frequency of 58.9 MHz and a 90^o pulse time of 5 µs at 40 W. The recycle delay time was fixed to 50 s. All ^6/7Li chemical shifts were referenced to solid LiF at -1.0 ppm. 2D ⁷Li-⁷Li exchange were acquired using the same 90^o pulse times as the 1D spectra. Additionally, 2D spectra were acquired using a recycle delay time of 3 s and a mixing time of 1.5 s.

Composite electrolyte samples were prepared for solid-state NMR with bidirectional galvanostatic cycling at 5-10 µA for 5-7 days with ⁶Li foil electrodes at room temperature. The cell was disassembled, the electrodes were removed, and any excess pieces of lithium metal were cut from the sample with a razor blade since residual metal can disrupt the magic-angle spinning. The ⁶Li-enriched samples were then loaded into 1.9 mm ZrO₂ rotors in ambient atmosphere; exposure to ambient is not expected to influence the distribution of Li⁺ in the sample.

Modeling

A detailed description of the computational model for estimating conductivity and transference number can be found in the Supplementary Materials.

Surface modification and stability of LLZO nanofillers

LLZO NPs and NFs (nanofillers generally) were modified via silane chemistry. Silane modification allows the covalent attachment of many pendant groups to the oxide surface, where these chemistries can increase the interaction with the surrounding polymer matrix to influence local Li⁺ transport. Since LLZO is very sensitive to humid air, we developed an anhydrous toluene-based synthesis procedure. LLZO nanofillers (1 g) were dispersed in 80 mL of toluene in an Ar-filled glovebox. The APTES silane reagent was then added to the mixture, which was stirred under reflux and nitrogen for 24 h. When a short reflux time was used (e.g., 4 h), Raman shifts confirmed that ATPES amount was not sufficient to protect LLZO from ambient exposure [Figure 1A and B]. The particles were then centrifuged and washed repeatedly to remove excess silane before final isolation and drying. The chemical composition of APTES-modified LLZO was characterized by ICP-OES [Supplementary Table 1]. Compared to the neat LLZO NP (Li = 5.99), the ATPES-modified LLZO NP shows slightly lower Li content (Li = 4.68), indicating loss of Li during the surface modification.

Figure 1. (A) Raman Zr-O peak in the LLZO nanofibers (B) Raman shift of Zr-O peak with air exposure time (C) FT-IR spectra of LLZO NP before (Neat LLZO NP) and after (APTES@LLZO NP) treatment with APTES (structure inset); (D) Differential absorbance spectra with APTES@LLZO NP subtracted from Neat LLZO NP. Positive values indicate new chemistries added to the material; negative values indicate chemistries removed from the material following silane treatment. LLZO: Li₇La₃Zr₂O₁₂; FT-IR: Fourier transform infrared spectroscopy; NPs: Nanoparticles; APTES: (3-aminopropyl)triethoxysilane.

Figure 1C shows FT-IR spectra of unmodified LLZO NPs (“Neat LLZO NP”) and LLZO NPs treated with an amine-terminated silane APTES (“APTES@LLZO NP”), showing clear emergence of peaks related to the silane (1,575 and 3,550 cm^-1 for N-H stretching) and attenuation of peaks (3,675 cm^-1 for O-H stretching) related to common LLZO surface contaminants. These changes are more apparent with differential absorbance [Figure 1D], where the appearance of N-H, C-H, and Si-O-Si bond vibrations show the APTES is successfully attached to the LLZO surface. Interestingly, there is also a loss of O-H and O₂C-O vibrations from LiOH and Li₂CO₃, respectively. This indicates that the treatment removed some (but still not all) of the native LLZO surface contaminant layer, something unexpected for a non-acidic, simplistic non-aqueous treatment^[26-29]. It is also possible that LiOH and Li₂CO₃ were prevented from forming further when the sample was exposed to air for the measurement.

This last observation reveals another benefit of silane modification of LLZO - the ability to process the electrolytes in air. To highlight this, we performed in situ Raman spectroscopy on LLZO NFs. The NFs have the same nominal chemical composition as the commercial NPs but have much higher surface area. This greater area increases the rate of reaction with humid air, leading to the formation of LiOH, which then reacts with CO₂ to form a Li₂CO₃ surface layer. Meanwhile, Li⁺ ions are exchanged for H⁺ in the bulk structure of LLZO [Figure 2A]^[30].

Figure 2. (A) Raman spectra of freshly annealed LLZO nanofibers from 3 to 246 min of air exposure as depicted on the right; (B-D) Detailed spectra showing peak shifts and intensity changes for (B) LLZO lattice vibrations; (C) Zr-O vibrations; and (D) C-O vibrations in Li₂CO₃; (E) Normalized integration of the Li₂CO₃ peak over time. Kinetics are estimated for pseudo-first-order reactions; (F) Spectra of the Zr-O band for LLZO NFs before and after 16 h of air exposure as well as 16 h of anhydrous treatment with GPS followed by 4 h of air exposure; (G) Comparison of the Zr-O band of freshly annealed LLZO NFs and LLZO NFs treated with GPS under acidified aqueous methanol and anhydrous toluene conditions. Raman shifts of the peak centers are labeled. LLZO: Li₇La₃Zr₂O₁₂; NFs: Nanofibers; GPS: Silane with terminal epoxide group.

Following heat treatment in the glovebox to remove carbonates, LLZO NFs were exposed to air with spectra taken after given time intervals. Even within a couple of minutes, the effects of carbonate formation are apparent. Figure 2B shows distortion and shifting of the cubic LLZO peaks as Li⁺ is replaced with H⁺ in the bulk structure. This process is most apparent in the Zr-O vibration at 640 cm^-1 [Figure 2C], which not only blue-shifts but forms a shoulder at higher wavenumbers as well^[31]. The reacted surface layer quickly accumulates Li₂CO₃, visible in Figure 2D. Interestingly, the accumulation of Li₂CO₃ seems to follow two distinct growth regimes [Figure 2E]: rapid growth in the first few hours of air exposure, followed by much slower but still continuous growth for up to a full day. Assuming a pseudo-first-order reaction, the kinetics for these growth regimes are 1.2 × 10^-2 s^-1 and 1.8 × 10^-4 s^-1, respectively. This may be due to the blocking of surface sites by Li₂CO₃ to hinder reaction with H₂O and/or the ability of the LLZO lattice to accommodate H⁺ and continue transporting Li⁺ to the surface^[32].

With the baseline LLZO response established, we examined the bulk LLZO lattice stability under certain conditions following silane treatment. For this study, we used a silane with terminal epoxide group (GPS) instead of APTES to prevent the silane from acting as a proton source or pH buffer, but the results should be generally applicable to many silane chemistries. Figure 2F shows the Zr-O band of LLZO NFs after 16 h of direct air exposure and after 16 h of reaction with GPS under anhydrous conditions. While there is still a shift in the Zr-O band (~9 cm^-1 blue-shift) indicative of some H⁺/Li⁺ exchange, there is much less than with unmodified LLZO exposed to air (~13 cm^-1 blue-shift). After silane treatment, exposure to air for several hours does not lead to any further shift of the peak, indicating the surface is passivated by the silane and any residual Li₂CO₃.

Another benefit of the anhydrous silane treatment is limiting LLZO lattice disruption during the reaction itself. Most silane treatments involve acidic, partially aqueous solutions to activate the silane chemistry and undergo quick reactions with the oxide surface^[16] However, as pointed out above, this is highly detrimental to LLZO. Figure 2G compares the Zr-O shift after GPS silane treatment and air exposure where the reaction was performed in anhydrous toluene or under an acidified methanol solution. The Zr-O blue-shift is significant (~19 cm^-1) and almost appears as two or three separate peaks for the acidic condition, indicating the extensive H⁺/Li⁺ exchange in the LLZO. This H⁺ replacement of Li⁺ is known to lower conductivity of bulk LLZO^[33,34] so it seems necessary to limit this as much as possible by processing in aprotic solvents^[34] to achieve optimal conductivity in LLZO-containing CPEs.

Conductivity of modified LLZO composites

To understand the effect of LLZO surface chemistry on Li⁺ transport, we fabricated CPEs with 0, 10, and 50 wt % LLZO NPs. The LLZO was either unmodified (“Neat”), modified with APTES via anhydrous toluene reaction (“APTES@LLZO NP”), or coated with ~10 nm of Al₂O₃ via ALD (“Al₂O₃@LLZO NP”), as shown in Figure 3A. In our previous work, we demonstrated that ~10 nm Al₂O₃@Cu electrodes have R_Int values exceeding 10 kΩ cm², indicating that these layers have very slow ion transport, as R_Int reflects the conductivity of ions (i.e., Li⁺) through the electrolyte-surface layers-electrode interfaces. Note that APTES acted as a model silane agent because it exhibits a slightly higher Li transference number compared to GPS-modified LLZO [Supplementary Figure 2]. Each of these represents control, positive, and negative conditions for how the LLZO surface hypothetically interacts with the surrounding polymer matrix without changing the particle morphology - covalently linking a polar organic linker to the surface should increase interaction with PEO more than the bare LLZO ceramic, while passivating the surface with a thick Al₂O₃ layer prevents any Li⁺ exchange between the two phases.

Figure 3. (A) Color scheme for modified LLZO nanoparticles; (B and C) Arrhenius-type plots of total bulk ionic conductivity for (B) 10 wt % and (C) 50 wt % LLZO NPs compared to PEO-LiTFSI (0 wt %); (D) Estimated transference numbers for all CPEs, quantified by the Bruce-Vincent current fraction ρ_BV. Small points represent values calculated from individual cells, and large points and error bars represent the mean and standard deviation of three cells; (E) Total ionic and Li⁺ conductivities as a function of LLZO NP content; (F) Apparent Arrhenius activation energies for all CPEs at temperatures above the melting point of PEO. LLZO: Li₇La₃Zr₂O₁₂; NPs: Nanoparticles; APTES: (3-aminopropyl)triethoxysilane; PEO-LiTFSI: Poly (ethylene oxide)-lithium bis(trifluoromethanesulfonyl)imide; CPEs: Composite polymer electrolytes.

At 10 wt % LLZO NPs, the total bulk ionic conductivities of Neat, APTES, and Al₂O₃ samples are incredibly similar and greater than that of PEO-LiTFSI alone at all temperatures [Figure 3B]. However, at 50 wt % LLZO NPs, the conductivities vary drastically. Neat LLZO is fairly similar to PEO-LiTFSI; APTES@LLZO demonstrates lower conductivity at most temperatures but with different temperature dependencies, and Al₂O₃@LLZO shows conductivities more than order of magnitude lower than all other samples [Figure 3C]. Clearly, surface chemistry has a greater effect at higher fractional contents in the composites. Note that in Figure 3, by Neat LLZO, the authors indicate CPEs where the surface of the LLZO ceramic NPs are not modified in any fashion.

The temperature-dependent ionic conductivities of different CPEs plotted in Figure 3B and C demonstrate a reverse “S”-type shape. The reported ionic conductivity values are very well within the range reported in existing literature, within 10^-8 - 10^-3 S/cm^[35] The flat region, with lower activation energy at higher temperatures (T > 55 °C), can be associated with faster ionic transport through the CPE with molten PEO. The higher activation energy observed at the intermediate temperatures (55 °C > T > 20 °C) can be attributed to the transport of the ionic species through the crystalline or semicrystalline PEO that demonstrates lower conductivity. The source of lower activation energy at lower temperatures (T < 20 °C) is unknown and can possibly be attributed to the variation in temperature dependence of ionic conductivity, which can be associated with the non-linear Vogel-Tammann-Fulcher (VTF) behavior. Note that transport of the lithium ions, which is considered to be the major ion-carrying species, is assumed to occur through the bulk of the PEO or PEO/LLZO interface, and not through the bulk of LLZO.

These numbers represent the total ionic conductivity, including contributions from both Li⁺ and the counterion TFSI^-. To probe true Li⁺ transport, t_Li+ values were estimated with the Bruce-Vincent method as current fractions (ρ_BV; see Supplementary Figures 3 and 4)^[25]. We note that this method - which is predicated on a homogeneous polymer system - is likely not applicable to truly quantify transference numbers for heterogenous composite electrolytes with highly variable impedances^[36]. Nonetheless, we expect that the qualitative trends here are still relevant based on the propensity of the electrolytes to achieve steady-state current-voltage profiles^[37] [Supplementary Figure 5] with triplicate measurements, even if the exact t_Li+ values may not be accurate. Figure 3D shows that adding 10 wt % LLZO NPs with any surface modification increases t_Li+ above that of PEO-LiTFSI, from ~0.13 to ~0.16 for Neat LLZO and APTES@LLZO (disregarding one outlier) and surprisingly to ~0.20 for Al₂O₃@LLZO. When moving to 50 wt %, the surface chemistries diverge in much the same way as in the case of bulk conductivity. APTES@LLZO exhibits the highest t_Li+ at ~0.21; Neat LLZO is close at ~0.19, but Al₂O₃@LLZO plummets to ~0.05. This agrees with the hypothesis that increasing LLZO-PEO interactions via silane chemistry helps Li⁺ transport, while coating with Al₂O₃ shuts down those interactions completely.

Figure 3E plots the total ionic conductivity and estimated Li⁺ conductivity (σ_Li+ = t_Li+ × σ_Total) as a function of LLZO NP content at 60 °C. At 10 wt % LLZO NPs, surface chemistry matters very little and increases in both total ionic conductivity and relative Li⁺ mobility are observed. At 50 wt %, Neat LLZO shows the highest Li⁺ conductivity, while APTES@LLZO shows similar σ_Li+ to PEO-LiTFSI despite having lower total conductivity. Al₂O₃@LLZO NPs at 50 wt % show extremely poor Li⁺ conductivity at the 10^-6 S/cm level. Despite these differences in total and Li⁺ conductivity, the apparent Arrhenius activation energy [Figure 3F] in the 55-70 °C range, above the melting point of PEO, does not vary significantly across all surface chemistries and LLZO weight fractions (all within the 0.44-0.56 eV range). This indicates that PEO still dominates ion transport in these systems, and that bulk LLZO is not directly involved in Li⁺ transport.

PEO-LLZO interphase resistance

The lack of through-LLZO Li⁺ transport (ion movement from the polymer phase into LLZO, through the bulk of the LLZO phase, and back into the polymer phase) is likely due to the high interfacial resistance between LLZO and the PEO-LiTFSI matrix^[38], which has been hypothesized to come from the restricted polymer movement, local polymer crystallization, and space charge effects at the ceramic interface^[9,12,13,39]. The fact that different Li⁺ transport is observed for both APTES@LLZO and Al₂O₃@LLZO compared to Neat LLZO implies that surface modification affects this process in some way, effectively changing the interphase resistance. We use the term “interphase” here because cations and/or anions (lithium, lanthanum, zirconium, carbon, oxygen, hydrogen, fluorine, sulfur, or nitrogen) must interdiffuse between an ordered ceramic phase and a more disordered polymer phase, and there is evidence in the literature that LLZO reacts with the ether chains in PEO to form decomposition layers more akin to a solid electrolyte interphase^[40].

To quantify this interphase resistance, trilayer cells consisting of a sintered LLZO pellet sandwiched between thin 10-15 µm PEO-LiTFSI layers with blocking electrodes were prepared [Figure 4A]^[24,41] The LLZO pellet was freshly annealed in an Ar-filled glovebox within 24 h of cell assembly to remove added resistance from surface contaminants. APTES@LLZO pellets were prepared by soaking the annealed pellet in a 3 wt % APTES solution inside the glovebox for five hours. From Nyquist plots alone, cells with APTES@LLZO pellets exhibit lower total impedance than those with Neat LLZO [Figure 4B and Supplementary Figure 6]. Fitting to a three-part equivalent circuit provided the best fits and shows that while the bulk conductivities (R1) of both cells are identical, the resistances of the medium frequency R2 and R3 elements are halved when the LLZO is modified with APTES [Figure 4C]^[42,43]. These medium frequency elements are attributed to the LLZO|PEO-LiTFSI interphase resistance, as LLZO and PEO-LiTFSI alone are only expected to show one R-CPE element due to bulk transport^[20]. We do not understand at this point why two semicircles are apparent (possibly due to resistance at the true microscopic LLZO|PEO interface followed by some longer-range resistance from the PEO domain, or due to imperfections in the LLZO pellet surface), but their combined resistance R2 + R3 is smaller with APTES modification regardless. The range of interfacial resistances observed in the trilayer analysis is around 200-500 W.cm², which resides within the range observed in other recent publications^[44,45].

Figure 4. (A) Schematic of the trilayer blocking cells; (B) Nyquist plots of PEO-LiTFSI|LLZO|PEO-LiTFSI cells at 60 °C with Neat LLZO (top) and APTES@LLZO (bottom) pellets. Grayscale lines show semicircles from each fit R-CPE element; (C) Equivalent circuit and fit resistances of the cells at 60 °C. Measurements at 25 °C can be found in Supplementary Figure 6. LLZO: Li₇La₃Zr₂O₁₂; PEO-LiTFSI: Poly (ethylene oxide)-lithium bis(trifluoromethanesulfonyl)imide; APTES: (3-aminopropyl)triethoxysilane; CPEs: Composite polymer electrolytes.

Lower interphase resistance for APTES@LLZO over Neat LLZO agrees with our observation of higher Li⁺ current fractions, but the decrease in total conductivity at 50 wt % APTES@LLZO NPs as compared to Neat LLZO NPs (as reported in Figure 3E) cannot be explained from the observations made using the trilayer configuration. This trend of maximum conductivity boost at 5-20 wt % ceramic filler followed by lower conductivity at higher weight fractions^[7,12] is theorized^[39] but unexplained in the field according to the existing literature, which will be attempted in the next couple of paragraphs using a theoretical model.

To fit this unintuitive conductivity trend, a PEO-LiTFSI-LLZO NP composite model was constructed with two interphases between the bulk polymer matrix and the LLZO NP. The outer interphase is treated as conductive (i.e., conductivity in this phase is higher than bulk PEO-LiTFSI; dark blue in Figure 5A), while the inner interphase is treated as resistive (conductivity is lower than bulk PEO-LiTFSI; dark orange in Figure 5A). Presence of the resistive interphase adjacent to LLZO is based on the experimental evidence of a decrease in ionic conductivity in polymer electrolytes near a ceramic surface due to the impeded polymer chain motion^[46]. The conductive interphase can be rationalized by the observation that adding nanofillers to semicrystalline polymers tends to decrease crystallinity^[47,48], so a locally amorphous layer of PEO around the LLZO domain would experience increased conductivity relative to the semicrystalline bulk^[49]. Interestingly, LLZO particles on the micron scale only decrease conductivity and do not show the local maximum at low weight fractions^[7], indicating this plasticizing might be both material- and size-dependent and requires nanofillers^[39]. The addition of this conductive interphase to a resistive interphase (already hypothesized theoretically^[12]), where both interphases are dominated by the polymer, is a new conception that will be explored here. The interphase resistance between the resistive interphase and LLZO was treated as 4 kΩ cm² in all cases, but previous modeling has shown that even the hundreds of Ω cm² of the APTES@LLZO|PEO-LiTFSI interphase at 60 °C is still orders of magnitude greater than the 1-10 Ω cm² required to achieve Li⁺ transport through the LLZO and not preferentially through the polymer phases^[38].

Figure 5. (A) Visualization of 10 wt % (4 vol %) and 50 wt % (20 vol %) LLZO NPs in a PEO-LiTFSI matrix. The bulk materials and interphases are labeled accordingly. Interphase resistances are not necessarily to scale; (B-D) Modeled values of LLZO NP composite electrolytes for (B) total ionic conductivity at 25 °C; (C) total ionic conductivity at 60 °C; and (D) effective bulk transference number at 60 °C overlaid with experimentally determined values. Fit parameters are found in Supplementary Table 1. LLZO: Li₇La₃Zr₂O₁₂; NPs: Nanoparticles; PEO-LiTFSI: Poly (ethylene oxide)-lithium bis(trifluoromethanesulfonyl)imide.

From Figure 5A, it is clear that at low LLZO content (10 wt % LLZO is roughly equivalent to 4 vol % LLZO in PEO-LiTFSI composites), the volume fraction of the composite comprised of conductive interphases is relatively high. A Li⁺ ion moving from one electrode to the other is likely to go through these regions, experiencing high mobility in the process. However, at high LLZO content, the volume fraction of resistive interphases takes over. In this case, there is essentially no pathway where ions can move without going through a resistive, low-mobility phase; or even if a percolating conductive phase exists, their volume fraction is usually very low and the pathway remains highly tortuous, which is unable to significantly contribute to the effective conduction of ions. The conductivities, thicknesses, and effective t_Li+ of these layers - especially the inner resistive interphase - may be dependent on the particle surface chemistry, which changes how the PEO interacts at that boundary. Interphase conductivity and thickness were varied for Neat LLZO, APTES@LLZO, and Al₂O₃@LLZO composites to identify best fits at 25 °C and 60 °C, as were transference numbers to fit experimental data at 60 °C [Supplementary Tables 2 and 3]. The interphase thicknesses ranged from 200-400 nm for the inner resistive interphase (slightly larger than the 70-190 nm thicknesses predicted by Bonilla et al. for higher molecular weight PEO chains^[12]), while the conductive interphase thickness ranged from 300-700 nm. Although this model treats the interphases as spatially distinct entities, in reality they should overlap to an extent due to the dynamic nature of the polymer phase. Also, considering the average size of the LLZO NPs being around 500 nm, the combined interphase regions with total thickness ranging between 500-950 nm substantially envelop the ceramic particles. Regardless, good fits between the model and experimental conductivity are seen in Figure 5B and C at low and high temperatures. Only incorporation of both conductive and resistive interphases allowed us to fit these trends, showing the complexity of the composite system.

It is worth pointing out that the mechanism of ion transport through the composite electrolyte is relatively simple, where the current is carried by the lithium cations (Li⁺) and the salt anions (TFSI^-, used in the polymer electrolyte) through the migration and diffusion processes. At the interface between PEO and LLZO, evolution of the space charge layer is not considered^[50], except for the presence of a large ohmic resistance that prevents transport of lithium ions from the polymer to the ceramic, and vice versa (for more details about the ion transport mechanisms and governing equations please see the Supplementary Information section). The experimentally observed increase and decrease in ionic conductivity with increasing ceramic fraction is captured by altering the properties of the polymer electrolyte located within a certain distance from the ceramic particles, which are characterized here as the resistive and conductive interphase regions. Ion transport mechanism within both the interphase layers constitutes migration and diffusion processes, which is very similar to the transport of ions within the bulk of PEO, but with different transport properties (such as conductivity, transference number, etc.). The flow of current within the composite electrolyte at room temperature and 60 °C is provided in Supplementary Figure 7 where it is clearly demonstrated that the majority of the current passes through the conductive interphase region.

The model is also able to replicate the macroscopic transference number by accounting for locally different transference numbers in the interphase regions [Figure 5D]. An effective bulk transference number was estimated from simulations and reflects a macroscopic value of microscopic interactions between bulk and interphase regions of differing transference numbers (t_Li+^Bulk and t_Li+^Interphase). While the bulk PEO-LiTFSI phase has similar transference numbers in all cases (t_Li+^Bulk ~ 0.11-0.18), the interphase controls effective Li⁺ mobility, with APTES@LLZO showing t_Li+^Interphase ~ 0.57 and Al₂O₃@LLZO showing essentially t_Li+^Interphase = 0 [Supplementary Tables 2 and 3]. This interphase-dependent Li⁺ mobility provides evidence for a surface-driven Li⁺ transport mechanism and further evidence that Li⁺ does not directly transport through the LLZO bulk in these CPEs to any appreciable degree. The physical-chemical phenomena behind these changes are still an open question but could be related to surface segregation of anions or cations at particle surfaces due to the surface polarity or H-bonding in the case of silane chemistries, followed by differences in PEO chain conformation at larger length scales around the LLZO^[12,39,51].

Chemistry and Li⁺ transport at the LLZO|PEO-LiTFSI interphase

Solid-state Li NMR was used to investigate the chemistry of the LLZO|PEO-LiTFSI interphase and experimentally demonstrate how modification influences Li⁺ transport at the interphase. Previous studies have used 2D NMR to measure Li⁺ exchange between ceramic and polymer phases in composites^{[8,14,40,52-55]} with some suggesting that secondary species such as LiOH on the LLZO surface mediate transport between these phases^[14].

⁷Li-⁷Li 2D exchange spectroscopy (EXSY) was conducted on 50 wt % Neat LLZO NP and APTES@LLZO NP composite electrolytes [Figure 6A and B] cycled at low current densities. Two peak regions with peaks show Li⁺ from the PEO-LiTFSI phase at -1.3 ppm and from LLZO phases at -0.6-1.0 ppm. The 2D NMR results show that for both Neat LLZO and APTES@LLZO composites, there is some overlap between both phases, indicating Li⁺ exchange from PEO to LLZO and LLZO to PEO. This demonstrates that Li⁺ is mobile to some extent through the composite interphase(s).

Figure 6. (A and B) ⁷Li-⁷Li 2D NMR spectra of (A) 50 wt % Neat LLZO NP and (B) 50 wt % APTES@LLZO NP composite membranes; (C) ⁷Li-¹H CPMAS spectra showing Li⁺-PEO interactions for both samples; (D and E) ⁶Li NMR spectra of the composite samples before and after ⁶Li||⁶Li symmetric cell cycling. LLZO: Li₇La₃Zr₂O₁₂; NPs: Nanoparticles; NMR: Nuclear magnetic resonance; APTES: (3-aminopropyl)triethoxysilane; CPMAS: Cross-polarization magic-angle spinning; PEO: Poly (ethylene oxide).

Notably, the LLZO region was very broad, with an upfield shoulder at -0.6 ppm discernable apart from the bulk LLZO signal at 0.1 ppm. This shoulder is typically attributed to the “LLZO-PEO interface”^[40,52,53], something that is vaguely defined and which has been shown above to be very complicated from an electrochemical standpoint. To interrogate the chemistry of this “interphase”, ⁷Li-¹H cross-polarization magic-angle spinning NMR (CPMAS) was carried out on both Neat LLZO and APTES@LLZO composites. As seen in Supplementary Figure 8, excitation of the ¹H nuclei in the PEO phase only results in a signal from the ⁷Li nuclei in the PEO-LiTFSI phase, not from the LLZO regions. Therefore, the chemistry of the “interphase” signal here is almost certainly LLZO-based. Whether this is a sub-population of non-bulk-like LLZO or a decomposed LLZO shell around the bulk LLZO particles^[53] is still in question but appears to be influenced by the surface coating [Supplementary Figure 7]. This “LLZO interphase” is not necessarily the resistive interphase used in our modeling [Figure 5] but could contribute to the resistance against Li⁺ transport through the bulk LLZO phase.

To better resolve and understand Li-dynamics in this interphase, 1D ⁶Li NMR was employed on Neat LLZO NPs and APTES@LLZO NPs. First, we note a decrease in Li signal for APTES@LLZO NP in comparison with Neat LLZO NPs [Figure 6C]. These results are in great agreement with Inductively Coupled Plasma Optical Emission Spectrometry (ICP-OES) measurements [Supplementary Tables 2 and 3]. The chemical composition of APTES-modified LLZO shows slightly lower Li content (Li = 4.68) compared to the neat LLZO NP (Li = 5.99), indicating loss of Li during the surface modification and Li-vacancy formation. Furthermore, a significant decrease in the linewidth (full width at half maximum, FWHM) for the LLZO signal was observed [Figure 6C]. The peak for APTES@LLZO exhibits a FWHM of 46 Hz, approximately 30 Hz smaller than that of the Neat LLZO NP (72 Hz). Generally, a smaller linewidth indicates ‘faster motion’ of the nuclei, or greater rotational degrees of freedom. These observations together imply that the silane coating is modifying Li environment at the interface and causing the Li to move faster in the APTES@LLZO NP as opposed to the Neat LLZO NP. Given this, we expect that the interphase region between the silane coating and the LLZO will be Li⁺ conductive and will allow for faster Li⁺ transport than in Neat LLZO. This hypothesis is supported by the model that predicted higher conductivity of the resistive region for APTES@LLZO than Neat LLZO (see Supplementary Tables 2 and 3). To corroborate these findings, ⁶Li NMR was used on composite membranes in the pristine form and after cycling with isotopically enriched ⁶Li||⁶Li symmetric cells. This allows an indirect view of where Li⁺ moves in the cell during cycling by comparing the relative ⁶Li amounts among PEO-LiTFSI, LLZO interphase, and bulk LLZO phases. Figure 6D and E shows ⁶Li NMR spectra for 50 wt % Neat LLZO NP and APTES@LLZO NP composites before and after ⁶Li enrichment through cycling. In both cases, the LiTFSI signal and LLZO interphase signal increase, while the bulk LLZO signal remains essentially unchanged. This confirms that Li⁺ does not transport through the bulk of the LLZO particles, and apparently only moves through this interphase region. However, this interphase enrichment is more pronounced for APTES@LLZO than Neat LLZO, with a 52% relative increase in ⁶Li signal with the APTES treatment vs. 13% relative increase without [Supplementary Figure 7]. This trend shows that APTES treatment can increase interfacial Li⁺ transport in these LLZO-PEO composites.

Effects on cell characteristics and performance

With a fundamental understanding of how LLZO surface modification affects Li⁺ transport in CPEs, we finally turned to Li||Li symmetric cell cycling to determine how modification affects cell performance. Nyquist plots of EIS for representative Neat LLZO NP, APTES@LLZO NP, and Al₂O₃@LLZO NP composites are shown in Figure 7A. Despite having similar bulk resistance values with 10 wt % LLZO, the medium frequency impedance R_Anode-Int (attributed to a combination of SEI resistance, charge transfer resistance, and other factors^[56-58] at the electrolyte-Li anode interface) are surprisingly different. For Neat LLZO, the R_Anode-Int decreases by half relative to PEO-LiTFSI; for APTES@LLZO, it is about the same; and for Al₂O₃@LLZO, it more than doubles. For 50 wt % LLZO, the R_Anode-Int of Neat LLZO increases drastically above the 0 and 10 wt % cases; APTES@LLZO only increases slightly more than 10 wt %; and Al₂O₃ increases by an order of magnitude. These values better reflect the bulk conductivity trends of Figure 3, but there appear to be some hidden relationships between bulk Li⁺ transport and interface impedance. Note that the charge transfer resistance at the Li|PEO-LiTFSI interface, measured at 60 °C, is around 300 Ω.cm², which is within the range reported by other recent publications that did not apply external pressure^[59]. Lower resistances at the Li|PEO-LiTFSI interface can be obtained by either operating at higher temperatures^[60], applying external pressure^[61], or through the adoption of complex electrode/cell fabrication procedures^[43].

Figure 7. (A) Nyquist plots of representative Li||Li symmetric cells comparing PEO-LiTFSI with 10 and 50 wt % Neat LLZO NP (left), APTES@LLZO NP (middle), and Al₂O₃@LLZO NP composites; (B) Representative galvanostatic cycling curves (10 min applied current, 40 min rest) for 50 wt % LLZO NP composites with different surface modifications; (C) Critical current density as a function of LLZO content. Small points represent values measured from individual cells, and large points and error bars represent the mean and standard deviation of three cells; (D) Correlation between CCD and total ionic and Li⁺ conductivities; (E) Correlation between CCD and R_Anode-Int; (F) Correlation between R_Anode-Int and total ionic and Li⁺ conductivities; All points in (D-F) are individual cells. Reported R² values are from linear regressions (solid and dashed lines) of the presented data. LLZO: Li₇La₃Zr₂O₁₂; NPs: Nanoparticles; PEO-LiTFSI: Poly (ethylene oxide)-lithium bis(trifluoromethanesulfonyl)imide; APTES: (3-aminopropyl)triethoxysilane; CCD: Critical current density.

Next, the Li||Li cells were galvanostatically cycled at increasing 10 µA/cm² intervals for ten minutes at a time, followed by extended rest periods, to determine the critical current density (CCD) at which the cell exhibits soft-shorts or complete short-circuiting [Figure 7B and Supplementary Figure 9]^[62]. This unidirectional, short pulse-long rest protocol was chosen to only allow time for voids in the stripped lithium working electrode to fill in and not lead to current focusing that would cause the cell to prematurely short^[63]. We also chose short times to allow better comparison with the PEO-LiTFSI control condition, as previous work from our group with continual plating and stripping shows that the inclusion of ceramic fillers artificially increases the CCD by creating tortuous pathways for dendrite growth and breaking up lithium filaments before they short^[10], a mechanism not possible with 0 wt % LLZO.

Figure 7C shows the CCD as a function of LLZO content for the three different LLZO surface modifications. The measured CCDs are in the range of 100 µA/cm² to 250 µA/cm² (or 0.1 mA/cm² to 0.25 mA/cm²), which are very well within the range of critical currents reported in the literature for PEO-based dry polymer electrolytes^[64,65]. Surprisingly, none of the composites have significantly higher CCDs than PEO-LiTFSI alone, and many cells even short at lower current densities. This indicates that something other than transference number - the typical direct link to improved cycling performance - is controlling cell performance, as Figure 7C bears little resemblance to Figure 3D. To establish what controls CCD in these cells, linear regressions between CCD and total ionic conductivity, Li⁺ conductivity, and R_Anode-Int were performed [Figure 7D and E]. Surprisingly again, the correlation between σ_Li+ and CCD is low (R²~0.106), while the CCD correlations with σ_Total (R²~0.645) and R_Anode-Int (R²~0.546) are stronger. Another interesting observation is that correlations are clear for high resistance data (low conductivity and high R_Anode-Int), but at low resistances, the correlations are less certain due to spread in the data. This makes sense if bulk mass transport (σ_Total/Li+) is the primary limiting factor if the conductivity is very low, but some other factor in the cell controls CCD even when the cell is not near the mass transport limit. Our observations of all the cells tested here show that R_Anode-Int is usually a good indicator of how quickly a cell will short, as high interface resistance possibly leads to current focusing during Li plating, which quickly leads to filament formation and shorting.

The best correlation between all studied parameters [Supplementary Figure 10] is actually between R_Anode-Int and σ_Li+ (R²~0.664), as seen in Figure 7F. In this way, σ_Li+, which is related to the Li⁺ diffusivity - a true descriptor of limiting current of an electrolyte^[66-68] - does influence the CCD. However, due to the practical issues associated with many electrochemical cell designs (uneven pressure in the cell stack, heterogeneous thickness of electrolytes, electrode roughness and purity, etc.) and the propensity of lithium metal to form uncontrolled filaments, Li⁺ diffusivity only indirectly controls the measured performance limit of the cell. Recent works have demonstrated ways to circumvent some of these challenges to probe performance even in coin cell formats^[68,69].

In addition to these practical design improvements, we would like to draw the field’s attention to the fundamental and intricate ways Li⁺ transport as a bulk property can influence the behavior at interfaces within the battery [Figure 8]. Lithium-ion transport is not simply increased by adding Li-containing particles; as we have demonstrated, Li⁺ does not move directly from the polymer through LLZO under these conditions due to large interfacial barriers. Multiple interfaces and interphases exist in an inherently heterogenous CPE. The interphases formed between ceramic particles and the polymer matrix dictate not only total ionic conductivity but also local conductivity of Li⁺ around the particle surface. Depending on the volume fraction of the ceramic filler, as well as the surface chemistry that may help or hinder Li⁺ mobility, the resulting size and conductivity of these interphases lead to global increases or decreases in Li⁺ conductivity. At the electrode interface, Li⁺ must move from the bulk electrolyte into the SEI, which is typically assumed to be a single-ion conductor made of inorganic and organic salt-like lithium compounds. Therefore, lithium-ion conductivity - a mixture of both transference number and total ionic conductivity - and not total conductivity dictates how quickly Li⁺ cations can make it to the anode surface to plate lithium, even if LLZO is not directly in or present at the SEI. In this way, “interface resistance” is not just a measure of how conductive the SEI is, but also how the bulk and the SEI interact. Since most lithium metal cells with composite electrolytes at the research level fail due to shorting from inhomogeneous lithium metal plating that is dependent on the SEI, it is important to determine how the fundamental, measurable properties of the cell - bulk, SEI, interface, and diffusion resistances - are linked to design better electrolytes moving forward.

Figure 8. Schematic showing how local interphases in the bulk composite electrolyte can influence resistance at the electrode-electrolyte interface.